Log in


7 May 2020 9:55 AM | Marie-Louise Rankin (Administrator)

The Statistical Society organised an online meeting of statisticians and epidemiologists to discuss the statistical issues concerning how Australia is battling COVID-19. Below are the five recommendations that resulted from the meeting.


IN THE SHORT TERM (i.e. over the next month)

1.        The tests to date have been on a ‘self-selected’ basis. They will not be representative of the population at large and so care needs to be taken on what inferences can be made about the population at large. For example, NSW and Victoria are encouraging testing to anyone who wants it. This will have a bias towards those people who have the time and capacity to drive to a site. Young adults and children are less likely to be tested. The better educated are more likely to be tested. Statisticians could provide advice on how get the best possible population inferences from the active virological (self-selected) testing regime (by geo-demographic reweighting?) and the likely limitations (biases). They could explain what the estimates mean (e.g. possible upper bound rather than actual ratio of infected persons). The extent of advice depends on the amount of geo-demographic information that is associated with individual test results.

2.     At first glance, the self-selected sample might seem to produce an upward bias because people who are symptomatic are more likely to be tested. However, it is also likely that the self-selected sample will under-represent children and young adults especially males. These may be more likely to be asymptomatic or mildly symptomatic although not tested and, if so, this would provide a bias in the opposite direction. Statisticians could provide advice on how to best supplement the active testing sample with populations that will be under-represented.

3.      Statisticians could review procedures for estimating R0 (the basic reproduction number). This is a crucial statistic for future planning but the statistical science behind it might be able to be improved (Methodology available on the Doherty web site). Also, there are a number of assumptions in the calculations such as all asymptomatic cases actually being pre-symptomatic. It would be importance to know the extent of uncertainty created by these assumptions through tools such as sensitivity analysis.

IN THE MEDIUM TERM (Until the virus is under control but should start work in the short term)

4.     As soon as feasible, a national survey be conducted jointly by the ABS and health authorities to estimate the number of infections after restrictions start being released. This information is very important for monitoring the potential for a second wave that is difficult to contain. It is may become more important in the winter months when there will be more overlap between COVID-19 symptoms and other winter ailments. Also, the number of asymptotic infections is unknown. International studies show the percentage is much higher than what health experts believe is the case in Australia. Statisticians might be able to assist the ABS with the analysis of the data. The survey should be repeated until there is confidence the virus is under control.

5.      Assist with the design, calibration and analysis of sewage testing. It could be combined with the national survey in the first instance to provide a ‘calibration experiment’. This would allow calibration with what we see in the population, reducing the need for large scale surveys in the future.

Dennis Trewin


  • 13 May 2020 9:08 AM | David Griffiths, retired Foundation Professor of St@tistics University of Wollongong
    It is of great importance that the Society and Statisticians involve themselves in addressing issues relating to COVID-19 testing. One other statistical method which deserves its hat to be thrown into the ring is pooled testing, introduced in a 1943 paper by Robert Dorfman and rapidly taken up by the US military to test for syphilis soldiers returning from the war.

    Since then, pooled testing has been used to good effect for various viral and bacterial diseases and in industrial applications. Successful small scale trials of pooled testing for COVID-19 have been reported in the media in several countries including Israel, Germany and USA. Other countries reported to have trialled this include New Zealand, Korea and India.

    Through personal correspondence with senior staff at the Doherty Institute. and a media report of work in the pathology labs at the Prince of Wales Hospital, it is clear that Australia has also undertaken small-scaled trials of pooled testing in its two most populous states

    Perhaps rather crudely, let me provide a simple description of two types of tests and their uses, as well as possible benefits of pooled testing for these two types of tests.

    Mucous membrane swabs test for active infection and return a positive through the detection of viral DNA. At both of our leading institutions that have trialled pooled testing this has been with small numbers (4 or 5) of individual samples pooled before testing. Of course, if any pooled test is positive, subsamples saved from individuals can be tested to find which members of a pool are infected. Because of the time delay and possible technical issues in sample splitting and RNA detection, pooled testing may gave a limited role in diagnostic testing.

    Blood samples test for prior infection (but not current infection in the first few days post-infection) through antibody detection This seems to provide an excellent vehicle in which to apply pooled testing to surveillance. In surveillance testing, finding the infection status of individuals may be of low priority, with the focus being on assessing the infection status of the population. But, of course, pooled testing allows testing of individual (sub)samples if desired. Pooled testing of blood samples, using large pool sizes might prove to be valuable in providing more information about the pool and about population infection rates than can be obtained by sewage surveillance mentioned in Denis Trewin’s report.

    Pooled testing has considerable potential value and we statisticians are the right people to provide advice on designing pooled tests, addressing technical matters such as optimal pool size, adaptive pooling and multi-stage pooling.
    Link  •  Reply
Powered by Wild Apricot Membership Software